Entropy-Based Machine Learning Model for Fast Diagnosis and Monitoring of Parkinson's Disease.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Parkinson's disease (PD) diagnosis and monitoring remain challenging, necessitating advanced analytical tools.
• Rest-state electroencephalogram (rs-EEG) signals offer a non-invasive window into brain activity relevant to neurological disorders.
• Current diagnostic methods may lack the efficiency and accessibility required for widespread, continuous patient monitoring.

Purpose of the Study:

• To develop a computationally efficient machine learning (ML) model for diagnosing and monitoring Parkinson's disease (PD).
• To evaluate the efficacy of different entropy calculation methods for PD detection using rs-EEG.
• To identify optimal signal characteristics (frequency range, brain hemisphere) and segment lengths for accurate PD classification.

Main Methods:

• Utilized rs-EEG signals from 20 PD subjects and 20 normal control (NC) subjects at a 128 Hz sampling rate.
• Compared various entropy calculation methods, identifying fuzzy entropy as the most effective for PD diagnosis.
• Implemented a feature selection procedure to reduce computational costs while maintaining classification accuracy.

Main Results:

• Fuzzy entropy demonstrated superior performance in diagnosing and monitoring PD using rs-EEG, achieving a classification accuracy (A_RKF) of approximately 99.9%.
• The most diagnostically relevant frequency range for PD was identified as 0-4 Hz, with informative signals predominantly from the right cerebral hemisphere.
• Classification accuracy significantly decreased with shorter rs-EEG segments (below 150 samples).
• Feature selection reduced computational costs by 11 times without compromising the ~99.9% classification accuracy.

Conclusions:

• A computationally efficient ML model utilizing fuzzy entropy on rs-EEG signals can accurately diagnose and monitor Parkinson's disease.
• The findings highlight the importance of specific frequency bands (0-4 Hz) and signal origins (right hemisphere) for PD detection.
• The proposed method is suitable for implementation in healthcare Internet of Things (H-IoT) applications, enabling low-power edge devices for enhanced PD management and patient resilience.